compiler may compile inline member functions as noinline blocks for different architectures, and linker may pick the newer arch, that results illegal instructions on old hardware

* quantize and dequantize tests

* unify activation and usability function

* drop NCNN_REQUANT cmake option, test dequantize requantize pack8, fix webassembly build

* benchmark use requantize int8 model

* feat(ncnnoptimize): replace denormals to zero on layers with weights

Co-authored-by: youngsoo.lee <youngsoo15.lee@gmail.com>
Co-authored-by: Restyled.io <commits@restyled.io>

* use local pool allocator

* detach extract feat from local allocator

* fix test

* faster bbox calculation by background score

* Restyled by clang-format

* Restyled by astyle

* Restyled by clang-format

* Restyled by astyle

Co-authored-by: Qoo <r97922153@gmail.com>
Co-authored-by: Restyled.io <commits@restyled.io>

* arm neon optimization for conv3x3s1 winograd42

* better condition

* Update test_convolution.cpp

* Update test_convolution.cpp

* more proper conditions

* arm neon optimization for general im2col sgemm pack4

* add sgemm

* wip

* wip

* fix armv7 build

* more conditions blah blah

* code format

* fix convolution

* move packed convolution to seperated header source

* unify weight data bf16

* proper conditions

* conv3x3s2 sgemm pack4 test

* feat: add denormal options

Flush-To-Zero(FTZ) and Denormals-Are-Zero(DAZ) are modes that bypass IEEE754 methods of dealing with denormal floating-point numbers on x86_64 and some x86 CPUs.

* feat: Integrate `flush_denormals` into `Extractor::extract`

* chore: replace global variable with `ThreadLocalStorage`